Combining traditional media mix modelling with attribution
Combining the best elements of media mix modelling and media attribution enables marketers to effectively plan, attribute, forecast and optimise their marketing activities.
The exponential growth of digital marketing has resulted in an increased focus on the impact and return of investment (ROI) of major marketing programs. As more and more analog channels such as TV and radio become digital, various marketing analytics burgeon to give companies insight into the levels of effectiveness of their various marketing activities. However, having multiple types of ROI measurement across these activities can be inefficient and ineffective.
We also know that an increasing amount of time is spent engaging in digital channels that lend themselves to very granular digital purchase path tracking and media attribution (MA). In addition to this, more and more media channels are becoming addressable. Accordingly, organisations that leverage these facts early on for measurement are going to be at an advantage. Marketing analytics increasingly utilises media attribution principles and the companies that build up experience with a combined approach early on will be well positioned for the future.
There are proven benefits of media attribution in optimising the media mix and improving performance, which was covered in the Econsultancy research paper, State of Marketing Attribution in APAC. Likewise, every media mix model (MMM) has its benefits as well as drawbacks as shown below. Hence, Datalicious came up with an approach that combines the best elements of MMM and MA to create a unified ROI currency that allows granular planning and forecasting. For many of our largest clients with big media budgets, the digital paths we have constructed for media attribution provide very wide coverage of the population as 92% of Australian adults are now online.
Although we cannot measure the effects of above the line (ATL) advertising (such as traditional TV and outdoor) directly in a deterministic sense, the effects of the behavioural changes it engenders can be detected in online behaviour.
This process is enabled by modelling ATL and external factors using traditional econometric models first. Datalicious’ integrated approach is customised to suit available data granularity as shown in the figures below.
Once ‘top-down’ ATL effects are known, the revenue generated can then be attributed at the most granular level, the individual, thanks to the existence of millions of digital paths. Traditional MMM, even with multi-layered models very often only allows visibility at a product / geo level at best. The figure below shows how the individual user level MMM paths allows multi-layered optimisation.
For ATL exposures, ‘touch-points’ can be inserted into these paths (in terms of a probability of seeing an ad) based in target audience research profiles matched to the customers profiles of individuals stored in the client’s back end customer databases. The figure below illustrates the different methodologies to inject offline (ATL) touchpoints into purchase paths.
The first methodology establishes correlations between offline (ATL) ads and online response spikes and is used to estimate probability of exposure for each user path. TARP matching is another method wherein if previous methods yield no results we fall back to matching audience profiles to media run and spend data at the most granular level available.
Agent-based modelling techniques like Markov Chains, combined with machine learning will then allow optimisation scenarios to be run at any combination of ‘what-if‘ factors the client wishes to look into. Geographical data can be combined with campaign, product and customer segment. Further integration of this technique with Markov modelling was found to be beneficial in describing, analysing and projecting results to allow planning at the most micro level.
Granularity allows machine learning specific to optimization parameters. The figure below illustrates how when new set of conditions are imposed, the model is re-run, not simply ‘sliced’. Marketers can derive another version of optimal spend with boundary conditions set by the business for any potential channel and medium combination. Granularity therefore allows comparison between the unconditional optimal mix and the boundary-conditioned optimal mix, thus addressing the frequent complaint that some of the scenarios thrown up by traditional MMM are unachievable or unrealistic.
Combining the best elements of media mix modelling and media attribution enables marketers to take advantage of both models, allowing marketers to effectively plan, attribute, forecast and optimise their marketing activities. Most importantly, this gives marketers the capacity to focus their budget on programs that provide the greatest return.
The full deck is also available on SlideShare.
Originally published at blog.datalicious.com on February 24, 2016.